9 . 1 Review of Dynamic Time

Dynamic Time Warping is a technique for aligning two sequences of symbols. For speech recognition, the obvious application is aligning a captured utterance with an idealized template utterance. In general the two sequences are not of equal length, and the expansion or compression needed to align the sequences may not be uniform. For example, a template utterance may be spoken at a uniform rate with no disfluencies, whereas an actual utterance may dwell on some phones, move rapidly through others, omit a few, insert some extraneous phones (e.g. “uhm”), and repeat certain sections, as in stuttering.